Spatiotemporal Models for Region of Interest Analyses of Functional Neuroimaging Data

Abstract

In vivo functional neuroimaging technology enables the evaluation of behavior-related changes in measured brain activity within specific cortical regions of interest (ROIs). When sufficient neurophysiologic evidence exists to restrict attention to a defined cortical region, an ROI analysis can provide powerful insights regarding neural representations of cognition, emotions, behaviors, and the neuropathology of psychiatric disorders. Given the complexity and abundance of data from neuroimaging experiments, anatomically focused research questions allow statisticians to explore models that more accurately reflect neurophysiologic characteristics of the data than global activation studies. Neural processing characteristics of particular interest in this article are spatial correlations stemming from the interplay between spatially distinct brain locations and temporal correlations between serial measures of brain activity. Despite the simplified data structure of ROI studies, challenges remain in modeling spatial correlations due to, for example, the fact that the correlations do not necessarily decrease as a function of increasing separation between the measurement locations. This article presents a spatiotemporal model that incorporates a functionally defined distance metric into a parametric structure for spatial correlations and includes temporal correlations between repeated scans. We demonstrate the use of the spatiotemporal model using experimental data from a study of the effects of ethanol administration on brain activity in the cerebellum, which largely controls balance and posture. We further illustrate our model using data simulated from a study evaluating neural processing alterations in the right prefrontal cortex associated with mental arithmetic.